Statistically unbiased prediction
enables accurate denoising of voltage imaging data
Nature Methods
- Minho Eom* KAIST
- Seungjae Han* KAIST
- Gyuri Kim KAIST
- Eun-Seo Cho KAIST
- Jueun Sim KAIST
- Pojeong Park Harvard University
- Kang-Han Lee Chungnam National University
- Seonghoon Kim Seoul National University
- Márton Rózsa Allen Institute
- Myunghwan Choi Seoul National University
- Cheol-Hee Kim Chungnam National University
- Adam E. Cohen Harvard University
- Jae-Byum Chang KAIST
- Young-Gyu Yoon KAIST
Abstract
Here we report SUPPORT (Statistically Unbiased Prediction utilizing sPatiOtempoRal information in imaging daTa), a self-supervised learning method for removing Poisson-Gaussian noise in voltage imaging data. SUPPORT is based on the insight that a pixel value in voltage imaging data is highly dependent on its spatially neighboring pixels in the same time frame, even when its temporally adjacent frames do not provide useful information for statistical prediction. Such spatiotemporal dependency is captured and utilized to accurately denoise voltage imaging data in which the existence of the action potential in a time frame cannot be inferred by the information in other frames. Through simulation and experiments, we show that SUPPORT enables precise denoising of voltage imaging data while preserving the underlying dynamics in the scene.
Citation
Website credit : Michaël Gharbi and Matt Tancik.